Skip to content

Dataset · yiting

Benchmark for safe and unsafe image content

yiting/UnsafeBench

10,000 real and AI-generated images labeled across 11 categories of unsafe content.

What's inside

Contains 10,146 images (6,098 safe and 4,048 unsafe) categorized across 11 types of unsafe content, along with text descriptions or prompts where available. The data is split roughly evenly between two sources: real photos from LAION-5B (5,060 images) and AI-generated images from Lexica (5,086 images).

What you can do with it

  • Training image moderation classifiers and filters (NSFW/safety)
  • Testing safety system robustness against both real and AI-generated images
Tasks
Image classification
Size
10K–100K rows
Downloads this month
739
Likes
41
Access
on request on Hugging Face

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("yiting/UnsafeBench")